ebook img

A Proof of George Andrews' and Dave Robbins' q-TSPP Conjecture (modulo a finite amount of routine calculations) PDF

0.14 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview A Proof of George Andrews' and Dave Robbins' q-TSPP Conjecture (modulo a finite amount of routine calculations)

A PROOF OF GEORGE ANDREWS’ AND DAVE ROBBINS’ q-TSPP CONJECTURE (MODULO A FINITE AMOUNT OF ROUTINE CALCULATIONS) MANUELKAUERS∗,CHRISTOPHKOUTSCHAN∗,ANDDORONZEILBERGER∗∗ Accompanied byMaple packages TSPP and qTSPPavailable from http://www.math.rutgers.edu/~zeilberg/mamarim/mamarimhtml/qtspp.html. 9 0 Pour Pierre Leroux, In Memoriam 0 2 n a Preface: Montr´eal, May 1985 J 7 In the historic conference Combinatoire E´num´erative [6] wonderfully organized by Gilbert La- 2 belle and Pierre Leroux there were many stimulating lectures, including a very interesting one by Pierre Leroux himself, who talked about his joint work with Xavier Viennot [7], on solving ] O differential equations combinatorially! During the problem session of that very same colloque, C chairedbyPierreLeroux,RichardStanleyraisedsomeintriguingproblemsabouttheenumeration . of plane partitions, that he later expanded into a fascinating article [9]. Most of these problems h concerned the enumeration of symmetry classes of plane partitions, that were discussed in more t a detail in another article of Stanley [10]. All of the conjectures in the latter article have since m been proved (see Dave Bressoud’s modern classic [2]), except one, that, so far, resisted the efforts [ of the greatest minds in enumerative combinatorics. It concerns the proof of an explicit formula for the q-enumerationof totally symmetric plane partitions, conjectured independently by George 2 v Andrews and Dave Robbins ([10], [9] (conj. 7), [2] (conj. 13)). In this tribute to Pierre Leroux, 1 we describe how to prove that last stronghold. 7 5 0 1. q-TSPP: The Last Surviving Conjecture About Plane Partitions . 8 Recall that a plane partition π is an array π = (π ), i,j 1, of positive integers π with finite 0 ij ≥ ij sum π = π , which is weakly decreasing in rows and columns. By stacking π unit cubes 8 ij ij | | 0 on top of thPe ij location, one gets the 3D Ferrers diagram, that can be identified with the plane- : partition,andis aleft-, up-, andbottom- justified structure ofunit cubes, andwecanreferto the v i locations (i,j,k) of the individual unit cubes. X A plane partition is totally symmetric iff whenever (i,j,k) is occupied (i.e. π k), it follows ij r ≥ a that all its (up to 5) permutations: (i,k,j),(j,i,k),(j,k,i),(k,i,j),(k,j,i) are also occupied. { } In 1995, John Stembridge [11] proved Ian Macdonald’s conjecture that the number of totally symmetric plane partitions (TSPPs) whose 3D Ferrersdiagramis bounded inside the cube [0,n]3 is given by the nice product-formula i+j+k 1 − . i+j+k 2 1≤i≤Yj≤k≤n − Ten years after Stembridge’s completely human-generated proof, George Andrews, Peter Paule, and Carsten Schneider [1] came up with a computer-assisted proof, that, however required lots of human ingenuity and ad hoc tricks, in addition to a considerable amount of computer time. ∗ SupportedinpartbytheAustrianScienceFoundation(FWF)grantsP19462-N18andP20162-N18. ∗∗ SupportedinpartbytheUnitedStates ofAmericaNationalScienceFoundation. 1 2 MANUELKAUERS,CHRISTOPHKOUTSCHAN,ANDDORONZEILBERGER Way back in the early-to-mid eighties (ca. 1983), George Andrews and Dave Robbins indepen- dently conjectured a q-analog of this formula, namely that the orbit-counting generating function ([2], p. 200, [9], p. 289) is given by 1 qi+j+k−1 − . 1 qi+j+k−2 1≤i≤Yj≤k≤n − In this article we will show how to prove this conjecture (modulo a finite amount of routine computer calculations that may be already feasible today [with great technical effort], but that would most likely be routinely checkable on a standard desktop in twenty years). 2. Soichi Okada’s Crucial Insight Our starting point is an elegant reduction, by Soichi Okada [8], of the q-TSPP statement, to the problem of evaluating a certain “innocent-looking” determinant. This is also listed as Con- jecture 46 (p. 42) in Christian Krattenthaler’s celebrated essay [5] on the art of determinant evaluation. Let, as usual, δ(α,β) be the Kronecker delta function (δ(α,β) = 1 when α = β and δ(α,β) = 0 when α=β), and let, also as usual, 6 a (1 qa)(1 qa−1) (1 qa−b+1) = − − ··· − . (cid:20)b(cid:21) (1 qb)(1 qb−1) (1 q) − − ··· − Define the discrete function a(i,j) by: i+j 2 i+j 1 a(i,j)=qi+j−1 − +q − +(1+qi)δ(i,j) δ(i,j+1). (cid:18)(cid:20) i 1 (cid:21) (cid:20) i (cid:21)(cid:19) − − Soichi Okada ([8], see also [5], Conj. 46) proved that the q-TSPP conjecture is true if 1 qi+j+k−1 2 det(a(i,j))1≤i,j≤n = (cid:18)1−qi+j+k−2(cid:19) . 1≤i≤Yj≤k≤n − So in order to prove the q-TSPP conjecture, all we need is to prove Okada’s conjectured determi- nant evaluation. 3. Certificates for Determinant Identities In [15], an empirical (yet fully rigorous!) approach is described to (symbolically!) evaluate de- terminants A(n) := det(a(i,j)) , where a(i,j) is a holonomic discrete function of i and j. 1≤i,j≤n Note that this is an approach, not a method! It is not guaranteed to always work (and probably usually doesn’t!). Let’s first describe this approach in more general terms, not just within the holonomic ansatz. Suppose that a(i,j) is given “explicitly” (as it sure is here), and we want to prove for all n 1 ≥ that det(a(i,j)) =Nice(n), 1≤i,j≤n for some explicit expression Nice(n) (as it sure is here). The approachis to pull out of the hat another “explicit” (possibly in a much broadersense of the word explicit) discrete function B(n,j), and then check the identities n (Soichi) B(n,j)a(i,j)=0, (1 i<n< ), ≤ ∞ Xj=1 (Normalization) B(n,n)=1, (1 n< ). ≤ ∞ A PROOF OF GEORGE ANDREWS’ AND DAVE ROBBINS’ q-TSPP CONJECTURE 3 If we could do that, then by uniqueness, it would follow that B(n,j) equals the co-factor of the (n,j) entry of the n n determinant divided by the (n 1) (n 1) determinant (that is the × − × − co-factor of the (n,n) entry in the n n determinant). Finally one has to check the identity × n (Okada) B(n,j)a(n,j)=Nice(n)/Nice(n 1) (1 n< ) − ≤ ∞ Xj=1 If the suggested function B(n,j) does satisfy (Soichi), (Normalization), and (Okada), then the determinant identity follows as a consequence. So in a sense, the explicit description of B(n,j) plays the role of a certificate for the determinant identity. 4. The q-Holonomic Ansatz In what sense might B(n,j) be explicit? In [15] the focus was on holonomic sequences, in the present situation we will work with q-holonomic sequences. A univariate sequence F(n) is called q-holonomic if it satisfies a linear recurrence of the form a (q,qn)F(n+r)+a (q,qn)F(n+r 1)+ +a (q,qn)F(n+1)+a (q,qn)F(n)=0 r r−1 1 0 − ··· where a ,...,a are certain polynomials. A key feature is that F(n) is uniquely determined by 0 r such a recurrence and the initial values F(1),...,F(r). It is therefore fair to accept recurrence plus initial values as an explicit description of the sequence F(n). A bivariate q-holonomic sequence F(n,m) is uniquely determined by a linear recurrence of the form a (q,qn,qm)F(n+r,m)+ +a (q,qn,qm)F(n+1,m)+a (q,qn,qm)F(n,m)=0 r 1 0 ··· where a ,...,a are certain rational functions and F(1,m),...,F(r,m) are q-holonomic as uni- 0 r variate sequences in m. This construction can be continued to discrete functions of any number of variables. Notethatwhile everyq-holonomicdiscretefunctioncanbe describedasabove,noteveryfunction thatisdescribedasabove,witharbitrary polynomialsa isnecessarilyholonomic(usuallyitisn’t!). r However there are efficient algorithms for deciding whether a candidate discrete function given as above is holonomic or not. One empirical way of doing this is to use the description to crank out many values, and then “guess” a pure recurrence with polynomial coefficients in the other variable, m, that can be routinely proved a posteriori. Just as holonomic sequences [13], q-holonomic sequences have a number of important properties. We recall the most important ones: (1) If F(n ,...,n ) and G(n ,...,n ) are (q-)holonomic, then so are the sequences 1 d 1 d F(n ,...,n )+G(n ,...,n ) and F(n ,...,n )G(n ,...,n ). 1 d 1 d 1 d 1 d A(q-)holonomicdescriptionofthesecanbecomputedalgorithmicallygiven(q-)holonomic descriptions of F and G. (2) If(q-)holonomicdescriptionsofsomesequencesF(n ,...,n )andG(n ,...,n )aregiven, 1 d 1 d then it can be decided algorithmically whether F =G. (3) If F(n ,...,n ) is (q-)holonomic, then 1 d ∞ G(n ,...,n )= F(n ,...,n ,k) 1 d−1 1 d−1 k=X−∞ is (q-)holonomic. A (q-)holonomic descriptionof G(n ,...,n ) can be computed algorithmically given 1 d−1 a (q-)holonomic descriptions of F(n ,...,n ). 1 d 4 MANUELKAUERS,CHRISTOPHKOUTSCHAN,ANDDORONZEILBERGER 5. The Computational Challenge DenotebyB′(n,j)the sequencedefinedby (Soichi) and(Normalization). Whycanweexpectthat B′(n,j) is q-holonomic? A priori there is no reason why it should be. We have to hope. And we cansystematicallysearch forapotentialq-holonomicdescriptionofB′(n,j). Ifwefindsomething, we have won, if not, we have lost, but there is always the hope that a further search, with larger parameters, would be successful. One can use (Soichi) and (Normalization) to compute the values B′(n,j) for, say, 1 j n 35 ≤ ≤ ≤ and then make an ansatz for a linear recurrence, say, of the form 10 7 4 c qαnqβj B′(n,j+γ)=0. α,β,γ γX=0(cid:16)βX=0αX=0 (cid:17) Foreachspecificchoiceofnandj,thisequationreducestoalinearequationfortheundetermined coefficientsc . Withdifferentchoicesofj andn,wecreateanoverdeterminedlinearsystemfor α,β,γ the c . Solutions of that system, if there are any, are with good probability valid recurrences α,β,γ for B′(n,j). So in principle, we just have to solve a linear system. But in practice, this is not as easy as it might seem. The values of B′(n,j) are rational functions in q, and so are the solutions c . α,β,γ A dense linear system over the rational functions with 440 unknowns cannot be solved directly with Gaussian elimination. The intermediate expression swell would quickly blow up the matrix coefficients to an astronomic size. Also the computation of 465 values of B′(n,j) via (Soichi) and (Normalization) is not entirely for free, because it too requires solving dense linear systems whose coefficients are rational functions in q. We solved the system using homomorphic images. In a first step, we computed the values of B′(n,j) with q set to 2, and reduced modulo the prime p := 231 1. This can be done quickly. − Also the linear system for the ansatz above can be solved quickly within the finite field with p elements. Itturnedoutthatthereisaonedimensionalsolutionspace. Atthispoint,thereisgood evidence that the B′(n,j) satisfy a recurrenceof the above form, but we do not know the explicit form of the coefficients c yet. Only their homomorphic images are known. α,β,γ In the homomorphic image, 110 of the 440 coefficients c are zero. We next refined the ansatz α,β,γ fortherecurrencebydiscardingthetermsqαnqβjB′(n,j+γ)forwhichc wasfoundtobezero α,β,γ in the homomorphic image. Next we repeated the computation of the B′(n,j) and the solution of the linear system for q = 3,4,5,6,...,150, always computing modulo p. The modular images were then combined via polynomial interpolation, rational function reconstruction, and rational number reconstruction [4] to coefficients which are rational functions in q over the integers. The resulting candidate recurrence has a number of remarkable features. (1) The recurrence was obtained as a solution of a dense overdetermined linear system. An artefact solution to an overdeterminedsystem appears only with very low probability. (2) The integercoefficients inthe rationalfunctions c do notexceed43inabsolutevalue. α,β,γ For an artefact solution, integers with absolute value up to √p 109 would be expected ≈ with very high probability. (3) The polynomials 7 4 c qαnqβj (γ =0,...,10) α,β,γ βX=0αX=0 factorize into low degree factors. For example, the leading coefficient of the recurrence (γ =10) factors as (qj+6 1)(qj+10+1)(qn qj+9)(qn qj+10)(qj+n+9 1)(qj+n+10 1). − − − − − For an artefact solution, it would be expected with very high probability that all the polynomials are irreducible. A PROOF OF GEORGE ANDREWS’ AND DAVE ROBBINS’ q-TSPP CONJECTURE 5 (4) The recurrence produces the correct terms of B′(n,j) for values 35 < n 200 for q = 2 ≤ and modulo p, although these terms were not used in the computation of the recurrence. For an artefact solution, this is expected to happen with very low probability only. (5) TherecurrenceproducesthecorrecttermsofB′(n,j)forsmallnandj ifq isleftsymbolic or set to a numeric value different from 2,3,...,100, For an artefact solution, this is expected to happen with very low probability only. We have not the slightest doubt that the recurrence we found is correct. For a rigorous proof, we can define (“pull out of the head”) a sequence B(n,j) by a q-holonomic description consisting of the recurrence we discovered and some suitable univariate recurrences and initial values that are easy to obtain. It is contained in the Maple package qTSPP accompanying this article. The much easier q = 1 case (that would give a new proof to the already proved Stembridge theorem) is contained in the Maple package TSPP. Proving that B(n,j) = B′(n,j) amounts to proving that B(n,j) satisfies (Soichi). (Equation (Normalization) is automatically satisfied.) Thanks to algorithms of Chyzak, Salvy, Takayama [3, 12], proving (Soichi) in principle reduces to finitely many routine calculations. Finally, (Okada) is also of the form A = B where both sides are q-holonomic. The left side is q- holonomicbecauseoftheclosureundermultiplicationanddefinite-summation, andtherightside, Nice(n)/Nice(n 1)isnotjust q-holonomic(a solutionofsome linearrecurrencewith polynomial − coefficients (in q,qn)) but in fact closed-form (the defining recurrence is first-order). Summarizing, we have found a certificate B(n,j) for Okada’s conjectured determinant identity 1 qi+j+k−1 2 det(a(i,j))= − . (cid:18)1 qi+j+k−2(cid:19) 1≤i≤Yj≤k≤n − It only remains to prove rigorously that our certificate really is a certificate. While such a proof can in principle be carried out automatically, we found that in practice, i.e., with the currently available algorithms, software, and hardware, it remains a computational challenge. Even for the case q = 1, we are at present unable to complete the necessary non-commutative Gr¨obner basis computations. 6. The Semi-Rigorous Shortcut We believe that even today, performing the computations for a rigorous proof is feasible, but it would require a huge technical effort. But why bother? First, if we wait for twenty more years, the availabilityofbetter algorithms,better software,andbetter hardwarewill probablyenable us to finish up these finitely many routine calculations with no sweat. Besides, since now we know for sure that a fully rigorous proof exists, do we really want to see it? It won’t give us any new insight. The beauty of the present approach is in the meta-insight, reducing the statement of the conjecture to a finite calculation. Furthermore, we know a priori that there exists an operator P(q,qn,N)(whereN istheshiftoperatorinn: Nf(n):=f(n+1))thatannihilatesthedifference ofthe leftandrightsidesof (Okada). IfthatoperatorhasorderL,say,thenacompletelyrigorous proof would be to check (Okada) for 1 n L. At present, we are unable to find P, and hence ≤ ≤ do not know the value of L. But it is very reasonable that L would be less than, say, 400, and checking the first 400 cases of (Okada) (and analogously for (Soichi)) is certainly doable (we did it for L = 100, and L = 400 for TSPP, but you are welcome to go further). These are done in procedures CheckqTSPPin the Maple package qTSPP,and CheckTSPPin the Maple package TSPP, respectively. The correspondinginput andoutput files canbe found inthe webpageofthis article mentionedabove. Asatechnicalaside,let’sconfessthatMaplerunningonourcomputerwasonly able to check (Soichi) and (Okada) for L 30, for symbolic q, but for random numerical choices ≤ of q it went up to L=100,andit is easyto see that with sufficiently many choices of numericalq for a given L, one can prove it for symbolic q. In 1993, Zeilberger [14] proposed the notion of semi-rigorous proof. At the time he didn’t have any natural examples. The present determinant evaluation, that was shown by Okada to imply 6 MANUELKAUERS,CHRISTOPHKOUTSCHAN,ANDDORONZEILBERGER a long-standing open problem in enumerative combinatorics, is an excellent example of a semi- rigorous proof that is (at least) as good as a rigorous proof. Let us conclude by promising that if any one is willing to pay us $107 (ten million US dollars), we will be more than glad to fill in the details. 7. Postscript This article was originally submitted to a special volume of the Seminaire Lotharingien de Com- binatoire(SLC)inmemoryofPierreLeroux. While theeditorsandrefereeswerewillingtoaccept our paper, they demanded that we change the title and “tone down” our claim that we have a proof (even modulo a finite amount of calculations). Since there is a “mathematical” possibility (as the French would put it) that our “proof plan” would not work out, in which case we have nothing. We agree that there is a positive probability that our proof would turn out to be wrong. But that probability is orders-of-magnitude smaller than the probability that the editors of SLC do not exist. After all they, along with all of us, may be characters in a dream of a giant, and we would all disappear once that giant wakes up. Sincewebelievethatourtitle isagoodone,andallourclaimsaresound,wedecidedtowithdraw the paper from SLC, and publish it in the much more enlightened journal The Personal Journal of Ekhad and Zeilberger, as well as in arxiv.org. However, as a concession to the human sentiments expressed by the editors and referees and for those impatient people who can’t wait twenty years, or cannot afford ten million US dollars, let us conclude with a sketch on how to hopefully make the present approach yield a fully rigorous proof with today’s software and hardware. The key is to take advantage of the special structure of the entries a(i,j), and not just consider them as yet-another-holonomic sequence. For the sake of simplicity, let’s consider the q =1 case. Analogous considerations apply to the q-case. When q =1, the matrix entry is: i+j 2 i+j 1 a(i,j)= − + − +2δ(i,j) δ(i,j+1). (cid:18) i 1 (cid:19) (cid:18) i (cid:19) − − Let’s write it as: a(i,j)=a′(i,j)+2δ(i,j) δ(i,j+1) , − where i+j 2 i+j 1 a′(i,j)= − + − . (cid:18) i 1 (cid:19) (cid:18) i (cid:19) − It is also helpful to define C(n,j) := B(n,n j). Note that C(n,j) is defined to be 0 for j < 0, − C(n,0)=1 and C(n,1) has a certain conjectured holonomic description as a sequence in n. At this point it may be fruitful to introduce the sequence of polynomials n f (x)= C(n,j)xj . n Xj=0 (it may be more efficient to let the sum range from j = 0 to j = n+2). The j-free recurrence for B(n,j) given in the package TSPP translates to a certain linear recurrence equation, in n, with polynomial coefficients in (n,x), for f (x), and the original N-free recurrence, translates to n a certain linear differential equation, in x, with polynomial coefficients in n and x. Just like all the classical orthogonal polynomials (Legendre, Laguerre, Hermite, Jacobi, etc.), except that the relevantequationsarenolongersecond-order,(andthe f (x) arenotorthogonal). Inparticular n { } the discrete-continuousfunction (n,x) f (x) has a full holonomic description in its arguments. n → Now both (Soichi)and (Okada) can be easily transcribed to certainsimply-stated constant-term identities in (n,x). A PROOF OF GEORGE ANDREWS’ AND DAVE ROBBINS’ q-TSPP CONJECTURE 7 Indeed, (Soichi) is equivalent to the constant-term identity x(2 x) f (x) CT − +2xi xi−1 n = 0 , (Soichi′) (cid:20)(cid:26)(1 x)i+1 − (cid:27) xn (cid:21) − where CT stands for “constant term”, i.e. “coefficient of x0”. Calling the left side L(n,i), note that we can induct on both n and i, so we would be done once we have an annihilating operator of the form P(N,I,n,i) (here N is shift in the discrete variable n, and I is shift in the discrete variable i). Calling the constant-termand F(n,i,x), this means that all we need is find an operator of the form P(N,I,n,i,x d ) annihilating F(n,i,x). It shouldn’t be too hard to get dx two operators annihilating F(n,i,x) out of the ones that we already have for f (x), and then n express them as P(n,i,x,N,I,x d ), Q(n,i,x,N,I,x d ), say, and use non-commutative Gr¨obner dx dx base eliminiation to eliminate x. This may be much more efficient than finding “pure” operators of the form P(N,n,i,x d ), or P(I,n,i,x d ), where we have to eliminate two “variables” in the dx dx non-commutative algebra of recurrence-differentialoperators C[n,i,x,N,I,x d ]. dx As for (Okada), it is equivalent to x(2 x) f (x) CT − +2xn xn−1 n = Nice(n)/Nice(n 1) . (Okada′) (cid:20)(cid:26)(1 x)n+1 − (cid:27) xn (cid:21) − − Itisverypossible,thatgoingviathis continuous-discreteroutewouldmaketheproblemtractable with today’s software and hardware, and we leave it as a challenge in case any of our readers is interestedenoughto convertour(currently)semi-rigorousproofintoafully rigorousproof,rather than wait patiently for twenty years. References [1] George E. Andrews, Peter Paule, and Carsten Schneider, Plane Partitions VI. Stembridge’s TSPP theorem, Adv.Appl.Math.34(2005), 709-739. [2] David M. Bressoud, “Proofs and Confirmations”, Math. Assoc. America and Cambridge University Press (1999). [3] Fr´ed´ericChyzakandBrunoSalvy,Non-commutativeeliminationinOrealgebrasprovesmultivariateidentities, J.ofSymbolicComputation 26(1998), 187-227. [4] JoachimvonzurGathenandJu¨rgenGerhard,Modern Computer Algebra,CambridgeUniversityPress,1999. [5] Christian Krattenthaler, Advanced Determinant Calculus, S´em. Lothar. Comb. 42 (1999), B42q. (“The An- drewsFestschrift”,D.FoataandG.-N.Han(eds.)) [6] G.LabelleandP.Leroux(editors),Combinatoire´enum´erative,LectureNotesinMathematics1234.Springer Verlag,NewYork. [7] PierreLeroux and G. X. Viennot, Combinatorial resolution of systems of differential equations, I. Ordinary differential equations. In Combinatoire ´enum´erative, ed. G. Labelle and P. Leroux. Lecture Notes in Mathe- matics1234,285-293.SpringerVerlag,NewYork. [8] Soichi Okada, On the generating functions for certain classes of plane partitions, J. Comb. Theory, Series A 53(1989), 1-23. [9] RichardStanley,Abaker’sdozenofconjecturesconcerningplanepartitions.InCombinatoire´enum´erative,ed. G.LabelleandP.Leroux.LectureNotesinMathematics 1234,285-293.SpringerVerlag,NewYork. [10] RichardStanley,Symmetriesof plane partitions,J.Comb.Theory,SeriesA43(1986), 103-113. [11] John Stembridge, The enumeration of totally symmetric plane partitions, Advances in Mathematics 111, 227-243(1995). [12] Nobuki Takayama, Gr¨obner basis, integration and transcendental functions. In “Proceedings of ISSAC’90” (1990), 152-156. [13] Doron Zeilberger, A Holonomic systems approach to special functions identities, J. of Computational and AppliedMath.32,321-368(1990). [14] DoronZeilberger,Theoremsforaprice: Tomorrow’ssemi-rigorousmathematicalculture,NoticesoftheAmer. Math.Soc.40 # 8,978-981(Oct.1993). Reprinted: Math.Intell.16,no.4,11-14(Fall1994). [15] DoronZeilberger,TheHOLONOMICANSATZII.AutomaticDISCOVERY(!) andPROOF(!!) ofHolonomic Determinant Evaluations, AnnalsofCombinatorics11(2007), 241-247. 8 MANUELKAUERS,CHRISTOPHKOUTSCHAN,ANDDORONZEILBERGER ManuelKauers,Research Institute forSymbolicComputation,J.KeplerUniversity Linz, Austria E-mail address: [email protected] ChristophKoutschan,ResearchInstituteforSymbolicComputation,J.KeplerUniversityLinz,Austria E-mail address: [email protected] Doron Zeilberger, Mathematics Department, Rutgers University (New Brunswick), Piscataway, NJ, USA. E-mail address: [email protected]

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.